In the planning for a targeted observation system, the targeted area is often identified based upon the information derived from sensitivity analysis of a numerical weather prediction model with respect to the uncertainties in the model's initial conditions. Numerical models are approximations of the real world. They are different in both physics and numerics. It has been useful to compare results from different models to understand what are in common among the models and what are their individual strength and weakness. A very important issue related to the applications of adjoint models in sensitivity analysis and targeted observations is how different the results of an adjoint model is from those of another adjoint model. It is both theoretically and practically meaningful to learn the common features and differences in the results from different adjoint modeling systems. Over the past few years, two different mesoscale adjoint modeling systems have been developed at NCAR: MM5 Adjoint Modeling system (MM5-4DVAR) and Mesoscale Adjoint Modeling System (MAMS). In this study, the two adjoint modeling systems are used to conduct a series of sensitivity analyses from the perspective of targeted observations for a marine cyclogenesis case. Our preliminary experiments revealed that the two adjoint modeling systems sometimes produced similar results, but sometimes they did not, depending on the response function used. Interpretation of the behaviors of the two adjoint modeling systems will be presented. Implications of our results in the applications of adjoint models in targeted observations and data assimilation will also be discussed.